Results

A total of 1986 studies were screened, 355 of these were assessed for eligibility and 330 were excluded (Figure S1). Reasons for exclusion included non-original studies (221 were narrative reviews/meta-analysis), studies in which outcomes were not relevant for the present objectives, and studies reporting results that were not suitable for inclusion in the present systematic review or meta-analysis. After a full-text review of the 405 studies, 75 studies were included (Figure S1).
Out of 75 included studies, 33 studies were related to the outer layer biodiversity [23 used dust samples and 10 used environmental biodiversity measures (e.g., species richness index, land-use gradient, and plant diversity)] and 42 studies were related to the inner layer biodiversity (2 studies characterized urine and skin microbiota diversity, 16 studies collected stool samples, and 25 characterize airway microbiota diversity) (Figure 1).
Characteristics of the 75 included studies are shown in Table S1 (and online supplement pp 6-8). Furthermore, 39 of the 75 included studies were classified as poor quality, 5 were classified as fair quality, and 31 were classified as good quality (Table S1). Study design limitations were mainly due to unadjusted potential confounders and/or self-reported outcome.
3.1. Outer layer biodiversity
The summary results on the association between outer layer biodiversity and asthma, wheezing, and allergic sensitization are shown in Table S2 (and online supplement p 7).

Studies included in meta-analysis

Shannon diversity index
Figure 2 presents the forest plot for the association between Shannon diversity index based on dust sampling techniques and the risk of asthma. The summary effect estimate indicated a protective effect, but it was not statistically significant (OR (95% CI) = 0.77 (0.55; 1.06), I2 = 72.4%, p =0.027) (Figure 2). The funnel plot (Figure S3) suggested an asymmetric pattern, but the Egger’s test for publication bias was not statistically significant (t=3.16,p =0.195).
Bacterial richness
The summary effect estimate (odds ratio) for the association between bacterial richness and asthma [OR (95% CI) = 0.74 (0.57; 0.96)] was consistent with the hypothesis that exposure to higher biodiversity, assessed as bacterial richness, has a protective effect on the development of asthma (Figure 3). There was considerable heterogeneity (I2=71.8%, p =0.003) across the studies (Figure 3). The funnel plot (Figure S4) was asymmetric, indicating some publication bias (Egger’s test: t=5.15, p =0.007).
3.2. Inner layer biodiversity
Table S2 summarises the evidence on the association between inner layer biodiversity and asthma, wheezing, and allergic sensitization (online supplement pp 7-8).

Studies included in meta-analysis

Shannon diversity index
The mean/median of the Shannon diversity index among individuals with and without asthma was reported in 7 studies. However, the study conducted by Park et al. 16 was excluded based on quality. The random effects model (n=6) provided a significantly increased standardized mean difference [SMD (95% CI) = 0.31 (0.14; 0.48)], indicating that the bacterial diversity was slightly higher among individuals with asthma (Figure 4). There was no significant heterogeneity between study-specific estimates (I2=0%, p =0.88) (Figure 4), but the sensitivity analysis including the study conducted by Park et al.16 increased heterogeneity up to 60% (Figure S6).
Bacterial richness
The mean/median of the bacterial richness/abundance among individuals with and without asthma was reported in 6 studies. The study conducted by Park et al. 16 was excluded because of low quality. The summary standardised mean difference (95% CI) from the random effects model (n=5) was 0.25 (0.06; 0.44) (Figure 5), indicating that bacterial richness/abundance was slightly higher among individuals with asthma. Consistent with Shannon diversity index, there was no significant heterogeneity between study-specific estimates (I2=0%, p =0.70) (Figure 5), but when including the study conducted by Park et al. 16 the heterogeneity increased to 52% (Figure S10).
The funnel plots (Figure S14 and S15) show an apparently asymmetrical pattern that may be indicative of publication bias. Despite this apparent asymmetry, Egger’s tests for publication bias were not statistically significant (t=-1.43, p =0.226 for studies on Shannon diversity index, and t=-0.45, p =0.680 for studies on bacterial richness), suggesting absence of publication bias.
The meta-analysis of 4 study-specific effect estimates, investigating associations between bacterial richness and asthma, showed no significant association between bacterial richness and asthma [OR (95% CI) = 1.14 (0.83; 1.56)] (Figure 6). The forest plot shown in Figure 6 demonstrates significant heterogeneity (I2=62.0%, p =0.048) across the studies. The funnel plot (Figure S16) suggested an asymmetric pattern; however, the Egger’s test indicated no publication bias (t=-0.26, p =0.819).